改进的BP神经网络在传动轴结构设计中的应用

Application of improved BP neural network in structural design of drive shaft

Shaffer函数定义域在[-10,10]区间内,BP神经网络拟合该函数训练时间长,且无法达到期望精度,说明BP神经网络拟合复杂非线性函数能力需改善。文章提出了一种改进的BP神经网络,先对网络的输入进行K-Means聚类,BP神经网络训练采用大规模节点,聚类输入分别激活部分节点进行训练,每组聚类使用不同的节点,通过子网络训练聚类样本,减少了网络拟合难度。经测试改进的BP神经网络达到了精度。最后,用改进的BP神经网络进行了轴径的最优计算。

The Shaffer function defines the domain in the range of [-10,10], BP neural network fitting function costs long training time, and can not achieve the desired accuracy, it showed that the ability of BP neural network to fit the complicated nonlinear functions needs to be improved. This paper presents an improved BP neural network, first, we should carry out K-Means clustering for the input of network, BP neural network is trained by large scale nodes, partial nodes are activated by clustering input to train, each cluster uses different nodes, by sub network to train clustering samples, it reduces network fitting difficulty. The test results show that improved BP neural network meets the precision. Finally, the axle diameter is calculated optimally by the improved BP neural network.